{
  "id": 426999,
  "title": "Ensemble decides the score?",
  "url": "/competitions/cafa-5-protein-function-prediction/discussion/426999",
  "author_name": "",
  "post_date": "2023-07-26T04:05:10.105628500Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hi,<br>\nI feel like no matter how you change the models and embeddings, the final score is being decided by the ensemble that you do at the end. Does anyone else feel that way?</p>",
  "messages": [
    {
      "id": "2359196",
      "postDate": "07/26/2023 04:05:10",
      "content": "<p>Hi,<br>\nI feel like no matter how you change the models and embeddings, the final score is being decided by the ensemble that you do at the end. Does anyone else feel that way?</p>",
      "rawMarkdown": "Hi,\nI feel like no matter how you change the models and embeddings, the final score is being decided by the ensemble that you do at the end. Does anyone else feel that way?",
      "votes": null
    },
    {
      "id": "2362415",
      "postDate": "07/28/2023 04:53:59",
      "content": "<p>Yes, but sometimes I can top up the score when I feel like I might add value somewhere.</p>",
      "rawMarkdown": "Yes, but sometimes I can top up the score when I feel like I might add value somewhere.",
      "votes": null
    },
    {
      "id": "2363017",
      "postDate": "07/28/2023 12:58:55",
      "content": "<p>how are you topping up the score?? I tried training larger models, adding regularization strategies on ProtBERT embeddings, but I still get the same LB score</p>",
      "rawMarkdown": "how are you topping up the score?? I tried training larger models, adding regularization strategies on ProtBERT embeddings, but I still get the same LB score",
      "votes": null
    },
    {
      "id": "2363139",
      "postDate": "07/28/2023 14:31:41",
      "content": "<p>Mb you using public notebook with constant score?</p>",
      "rawMarkdown": "Mb you using public notebook with constant score?",
      "votes": null
    },
    {
      "id": "2363518",
      "postDate": "07/28/2023 18:08:21",
      "content": "<p>I'm a beginner, and this is my first Kaggle challenge. Nevertheless, I believe that using publicly available embeddings, or deriving new ideas based on these embeddings, could inherently keep the score at a same level.</p>",
      "rawMarkdown": "I'm a beginner, and this is my first Kaggle challenge. Nevertheless, I believe that using publicly available embeddings, or deriving new ideas based on these embeddings, could inherently keep the score at a same level.",
      "votes": null
    },
    {
      "id": "2375611",
      "postDate": "08/05/2023 18:54:42",
      "content": "<p>If you are referring to the top score public notebook refer to this discussion - <a href=\"https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/427002\" target=\"_blank\">discussion#427002</a></p>",
      "rawMarkdown": "If you are referring to the top score public notebook refer to this discussion - [discussion#427002](https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/427002)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2362415,
      "author_name": "yuichiroyoshida",
      "author_url": "",
      "post_date": "07/28/2023 04:53:59",
      "content": "<p>Yes, but sometimes I can top up the score when I feel like I might add value somewhere.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2363017,
          "author_name": "abhranta",
          "author_url": "",
          "post_date": "07/28/2023 12:58:55",
          "content": "<p>how are you topping up the score?? I tried training larger models, adding regularization strategies on ProtBERT embeddings, but I still get the same LB score</p>",
          "votes": null,
          "replies": [
            {
              "id": 2363139,
              "author_name": "bejeweled",
              "author_url": "",
              "post_date": "07/28/2023 14:31:41",
              "content": "<p>Mb you using public notebook with constant score?</p>",
              "votes": null,
              "replies": []
            },
            {
              "id": 2363518,
              "author_name": "yuichiroyoshida",
              "author_url": "",
              "post_date": "07/28/2023 18:08:21",
              "content": "<p>I'm a beginner, and this is my first Kaggle challenge. Nevertheless, I believe that using publicly available embeddings, or deriving new ideas based on these embeddings, could inherently keep the score at a same level.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2375611,
      "author_name": "priyanagda",
      "author_url": "",
      "post_date": "08/05/2023 18:54:42",
      "content": "<p>If you are referring to the top score public notebook refer to this discussion - <a href=\"https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/427002\" target=\"_blank\">discussion#427002</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2359196": "Hi,\nI feel like no matter how you change the models and embeddings, the final score is being decided by the ensemble that you do at the end. Does anyone else feel that way?",
    "2362415": "Yes, but sometimes I can top up the score when I feel like I might add value somewhere.",
    "2363017": "how are you topping up the score?? I tried training larger models, adding regularization strategies on ProtBERT embeddings, but I still get the same LB score",
    "2363139": "Mb you using public notebook with constant score?",
    "2363518": "I'm a beginner, and this is my first Kaggle challenge. Nevertheless, I believe that using publicly available embeddings, or deriving new ideas based on these embeddings, could inherently keep the score at a same level.",
    "2375611": "If you are referring to the top score public notebook refer to this discussion - [discussion#427002](https://www.kaggle.com/competitions/cafa-5-protein-function-prediction/discussion/427002)"
  },
  "source": "meta"
}